feat(disable-power-models): Update Other and Total Energy Metric calculations

If DisablePowerModels is enabled, any metrics with source "trained_power_model" that
are used in Other and Total Energy Calculations (ex. kepler_node_other_joules_total,
kepler_process_other_joules_total, kepler_container_other_joules_total,
kepler_container_joules_total, kepler_process_joules_total) are removed from the
calculation (by replacing the metric value with 0). Negative values will be replaced with
0. If all metrics used for Other and Total Energy Calculations are from source "trained_power_model",
then the Other and Total Energy Calculations will export 0 (they will not be removed from prometheus).

Signed-off-by: Kaiyi Liu <[email protected]>

Author: KaiyiLiu1234

pkg/metrics/metricfactory/metric_factory.go

@@ -46,6 +46,8 @@ func EnergyMetricsPromDesc(context string) (descriptions map[string]*prometheus.
 		} else if components.IsSystemCollectionSupported() && strings.Contains(allComponents, name) {
 			// TODO: need to update condition when we have more type of energy metric such as network, disk.
 			source = components.GetSourceName()
+		} else if strings.Contains(name, config.OTHER) {
+			source = ""
 		}
 		if !config.DisablePowerModels() || source != modeltypes.TrainedPowerModelSource {
 			descriptions[name] = energyMetricsPromDesc(context, name, source)

pkg/model/node_component_energy.go

@@ -109,10 +109,20 @@ func UpdateNodeComponentIdleEnergy(nodeMetrics *stats.NodeStats) {
 
 func addEnergy(nodeMetrics *stats.NodeStats, metrics []string, isIdle bool) {
 	for socket, power := range GetNodeComponentPowers(nodeMetrics, isIdle) {
+		corePower := power.Core
+		dramPower := power.DRAM
+		uncorePower := power.Uncore
+		pkgPower := power.Pkg
+		if config.DisablePowerModels() {
+			corePower = 0
+			dramPower = 0
+			uncorePower = 0
+			pkgPower = 0
+		}
 		strID := fmt.Sprintf("%d", socket)
-		nodeMetrics.EnergyUsage[metrics[0]].SetDeltaStat(strID, power.Core*config.SamplePeriodSec())
-		nodeMetrics.EnergyUsage[metrics[1]].SetDeltaStat(strID, power.DRAM*config.SamplePeriodSec())
-		nodeMetrics.EnergyUsage[metrics[2]].SetDeltaStat(strID, power.Uncore*config.SamplePeriodSec())
-		nodeMetrics.EnergyUsage[metrics[3]].SetDeltaStat(strID, power.Pkg*config.SamplePeriodSec())
+		nodeMetrics.EnergyUsage[metrics[0]].SetDeltaStat(strID, corePower*config.SamplePeriodSec())
+		nodeMetrics.EnergyUsage[metrics[1]].SetDeltaStat(strID, dramPower*config.SamplePeriodSec())
+		nodeMetrics.EnergyUsage[metrics[2]].SetDeltaStat(strID, uncorePower*config.SamplePeriodSec())
+		nodeMetrics.EnergyUsage[metrics[3]].SetDeltaStat(strID, pkgPower*config.SamplePeriodSec())
 	}
 }

pkg/model/node_platform_energy.go

@@ -95,6 +95,9 @@ func GetNodePlatformPower(nodeMetrics *stats.NodeStats, isIdlePower bool) (platf
 func UpdateNodePlatformEnergy(nodeMetrics *stats.NodeStats) {
 	platformPower := GetNodePlatformPower(nodeMetrics, absPower)
 	for sourceID, power := range platformPower {
+		if config.DisablePowerModels() {
+			power = 0
+		}
 		nodeMetrics.EnergyUsage[config.AbsEnergyInPlatform].SetDeltaStat(sourceID, power*config.SamplePeriodSec())
 	}
 }
@@ -103,6 +106,9 @@ func UpdateNodePlatformEnergy(nodeMetrics *stats.NodeStats) {
 func UpdateNodePlatformIdleEnergy(nodeMetrics *stats.NodeStats) {
 	platformPower := GetNodePlatformPower(nodeMetrics, idlePower)
 	for sourceID, power := range platformPower {
+		if config.DisablePowerModels() {
+			power = 0
+		}
 		nodeMetrics.EnergyUsage[config.IdleEnergyInPlatform].SetDeltaStat(sourceID, power*config.SamplePeriodSec())
 	}
 }